How to use TensorFlow on CS Lab Machines

You probably want to use a virtual environment to install and run tensorflow on the CS machines.

When you install software in the virtual environment, make sure you install the tensorflow-gpu package if you want to use the GPU. If you install both tensorflow and tensorflow-gpu, it will default to not using the GPU.

You may see the following error messages when running tensorflow:

Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA

That one just means you didn’t compile TensorFlow from source, and it might not run as fast as is possible. Probably harmless, but worth looking into if you intend to run lots of long jobs.

Ignoring visible gpu device ...
The minimum required Cuda capability is 3.5.

This means the machine you are running on doesn’t have a good enough GPU. See our machine specs page for info on which GPU each machine has.

You can also run the deviceQuery command to see data on your current machine’s GPU.

See also:

Back to SwatCS Help Docs